Time series analysis in critical care monitoring.

Time series analysis techniques facilitate statistical analysis of variables in the course of time. Continuous monitoring of the critically ill offers an especially wide range of applications. Several studies from different work groups show that autoregression, integration, moving average (ARIMA) models help to identify pathologic outliers and trends in physiologic variables in surgical critical care. The effect of therapeutic interventions on physiologic target variables has been estimated with interrupted ARIMA models. The time series before the therapeutic intervention were compared to changes under intervention using the same model including an intervention regressor. In most patients clinically relevant therapeutic effects could be statistically identified. Similarly, noneffective therapeutic maneuvers could be detected early, and eventually changes in therapeutic strategy initiated. These techniques appear to be most appropriate with electronic online measurements at short time intervals, e.g., heart rate, invasive pressures, regional oxygenation. But even on the basis of short time series of critical care monitoring variables, ARIMA models can successfully be employed for the analysis of laboratory variables and of therapeutic interventions. Nevertheless, due to high demands for manpower and to statistical methodological limitations, the general use of this methodology in clinical practice apart from controlled clinical studies cannot be recommended today. Nevertheless, time series analysis techniques bear a great potential for clinical applications. Ongoing studies will in the future allow us to apply time series analyses to a wide group of clinical problems. In clinical practice, time series analyses support a more analytical and reproducible approach toward the evaluation of pathologic changes and therapeutic effects in the individual patient. Present research focuses on the development of automatic methods for time series analysis that allow instantaneous statistical analysis at the bedside and algorithms for multivariate time series analysis. This would offer an option to the healthcare professional for a more reliable evaluation of the individual treatment. Therefore, it appears rewarding to invest further efforts into the development of medical time series analysis techniques.